16 When Eugenic Enhancement Meets the Myth of Genetic Reductionism

1. What Reductionism Is, and Isn’t

Reductionism is the doctrine that laws and facts at one level can be explained by (and, in that sense, reduced to) laws and facts at another more fundamental level (Nagel 1961; Sarkar 1998). Molecular biology since the mid-twentieth century provides many interesting examples. For instance, when red blood cells absorb oxygen using hemoglobin molecules, the presence of a little oxygen leads to more rapid absorption until the cells begin to become saturated. After that, the rate of absorption levels off. The process leads to an S-shaped absorption curve. This phenomenon, known as the Bohr effect, is an example of what is called cooperative behavior in which parts of a system interact with each other to perform a task better than what would have been accomplished in the absence of interaction. In the 1960s, this phenomenon at the cellular level was explained by the allostery model, which accounted for the Bohr effect based on the physical interactions between the four amino acid residue chains that make up hemoglobin macromolecules. According to the model, the four chains changed shape once the first oxygen molecule attached to them. The changed shape allowed more oxygen molecules to attach more easily (Monod et al. 1963). Thus, a phenomenon from cellular biology was explained at the level of the physics of macromolecules.

To take another example, Escherichia coli bacteria only produce an enzyme to digest lactose when there is lactose available in its environment. When lactose becomes unavailable, it turns off the production of that enzyme. This goal-directed (or adaptive) behavior was explained by Monod and Jacob (1961) in the late 1950s using a molecular operon model that was regarded as one of the major successes of the emerging field of molecular biology. The lactose molecule interacted with a repressor molecule attached to DNA close to the gene for the enzyme; because of this interaction, the repressor molecule got detached from the DNA and the genes next to it could begin to express the enzyme that digested lactose. The process could be initiated by a molecule shaped like lactose but was not digested by the enzyme: There was nothing mysteriously goal-directed about it. Cellular biology was thus again reduced to macromolecular physics. Indeed, some philosophers have viewed the emergence molecular biology as a triumph of such physical reductionism (Sarkar 1998).

Physical reductionism must be contrasted with a related but fundamentally different research program in twentieth-century biology, that of genetic reductionism (Sarkar 2002, 2004). The motivation for this program was the belief that genes were the most important determinants of all aspects of an organism’s structure and behavior, that is, all its phenotypic traits. In the 1920s, this view seemed plausible enough when geneticists were systematically showing that a large number of traits in every organism that was studied were inherited as predicted by Mendel’s laws of (transmission) genetics. In particular, in the famous fly room at Columbia University, Thomas Hunt Morgan and his students showed Mendelian patterns in the inheritance of hundreds of traits in the fruit fly Drosophila melanogaster (Sarkar 1998). On the basis of these observations, it was inferred that genes were causally responsible for the presence of these traits even though, unlike the situation of physical reductionism in molecular biology, the causal pathways from genes to traits (that is the genotype-phenotype map) remained unelucidated.

Genetic reductionism is the program that lies behind the claims of those who are willing to claim genetic etiology for a trait even without any knowledge of the causal pathways from the DNA molecules that form the genes to the trait itself. Genetic reductionism does not concern itself with the molecular mechanisms by which a trait emerges in an embryo. In the heyday of genetic reductionism in the 1920s, genes were believed to be constituted out of protein molecules rather than DNA: That is how far genetic reductionism has always been from the exploration of causes. This is not to suggest that genes do not play an important causal role in the emergence of traits. Rather, it is to underscore the fragility, both historically and today, of any assumption that genes are the sole (or even the most) important causal determinants of all traits.

When the HGP was envisioned and promoted in the biological community in the 1980s, critics pointed out the pervasive extent to which it was based on the assumption of genetic reductionism (Tauber and Sarkar 1992, 1993; Sarkar 1998). The HGP was supposed to deliver a complete human DNA sequence, and eventually a unique one for each individual. This sequence would provide a person’s genetic profile (assuming that parts of the sequence that are genes can be distinguished junk DNA). Now, the expectation was that a knowledge of these profiles would revolutionize biology by accounting for the genesis of form and function and also medicine by personalizing medical intervention. What genomics has shown is that the promised revolution of the HGP must be postponed, perhaps indefinitely (Hall 2010). The reason for this is the complexity of the genotype-phenotype map. Phenotypic traits—of medical relevance or otherwise—can only very rarely be predicted on the basis of DNA sequence data alone.

To put it bluntly, genetic reductionism is a failed doctrine that should be relegated to the dustbin of history. There are a few hundred human traits that are largely determined by one or two genes. These are well known because they have been easy to study using genetical experiments and because geneticists in the twentieth century were always looking for them because they are easy to study. (The same situation holds for other species.) But, once attention shifts to other traits, typically ones that are interesting for ordinary reasons, including all complex physical or behavioral traits, a large number of genes are involved in some capacity or other and they interact with an array of nongenetic factors. Moreover, the developmental history of these interactions matters.

Contrary to the expectations on which the HGP was founded, the genomics research spawned by the HGP has only demonstrated the limitations of genetic reductionism. Knowing genetic profiles provides very little insight into the functioning of organism and is only very occasionally relevant to medicine; we do not get very far by gazing at sequences. The rise of epigenetics is one of the understandable results of recognizing the limitations of genetics; if genes explain very little of the biology of the organism, it is time to move to other molecules present on chromosomes or elsewhere in the cell. Genes alone do not make an organism.

Rather, a mature organism is the result of a sequence of developmental interactions between an embryo, starting as a single cell, and its environment. This view of life is called contextual developmental construction. In sexually reproducing organisms, within the fertilized cell, expression of genes depends on a variety of epigenetic factors. This is true of all organisms, and not just humans. (In asexual organisms there is a single cell that starts the reproductive cycle by initiating development.) The physical environment matters tremendously, and not just because proper development requires the presence of water and a suite of chemical nutrients. Sometimes, even gravity matters. In experiments first carried out in the 1970s and 1980s, developmental biologists showed that, if chick or frog embryos are tilted to an incorrect angle in early development, the body does not have the normal shape and symmetry (Ruden et al. 2018).

For some traits, notably sex determination, temperature is the most important variable in many vertebrate species. For instance, the sex of most turtles and all crocodilians (alligators, crocodiles, caimans) is determined by the temperature at which eggs are incubated during a certain period of development. Typically, only a very small range of temperatures allows the formation of both males and females. Above (or below) that range only one of the sexes is formed. In most cases, lower temperatures lead to males and higher ones to females (Bull 1983). But temperature is only one of the environmental determinants seen for sex in animals. In Gannarus duebeni, a tiny amphipod (a type of crustacean), sex is determined by the time of exposure to sunlight: Young that are born earlier in the season turn mostly male, the later-born mostly turn female (McCabe and Dunn 1997). Perhaps even stranger is the case of the worm, Bonellia viridis. If a larva settles on the seafloor, it becomes female; however, if it is ingested by a female (of the same species) it migrates to the uterus and becomes male (Leutert 1975).

Temperature-dependent sex determination is an example of phenotypic plasticity: when organisms with the same genotype develop different phenotypes in different environments. This aspect of development was recognized shortly after the beginnings of genetics in the early twentieth century (Sarkar 1999). Phenotypic plasticity is ubiquitous. The European map butterfly, Araschnia levana, has a spring form that is bright orange with black spots and a summer form that is almost black with a white band. The forms are so different that, in the eighteenth century, Linnaeus classified them as different species (van der Weele 1995). The water flea, Daphnia cucullata, grow their “helmets” (parts of their head) to twice the normal size in the presence of predators (Agrawal et al. 1999). The induced change is inherited into the next generation. In humans, phenotypic plasticity is shown (in a perhaps trivial sense) by every cultural trait when there can be variation independent of genetic changes. It has also been interpreted as the capacity of cells to change their behavior in response to internal or external environmental cues (Feinberg 2007, 433).

Phenotypic plasticity shows the extent to which development depends on environmental context. It shows how organismic forms—structure and behavior—are differently constructed in different environmental contexts using genetic resources in different ways. If a single gene is being analyzed, its environmental context consists not only of extra-genetic factors but also of the presence of other genes in that organism’s genome. Phenotypic plasticity is the opposite of the type of gene specificity that successful germline editing is predicated upon. The existence of potential phenotypic plasticity underscores the importance of ensuring specificity before embarking on a program of gene editing to achieve desired phenotypic changes.

2. Dreams of Enhancement: Physical Traits

Contemporary proponents of genetic enhancement focus their attention on a small set of physical and behavioral traits which, they claim, should be encouraged to spread in the general population. Indeed, some of them indignantly proclaim that it would be ethically unpalatable for us not to enhance such traits in our children (Robertson 1994; Savulescu 2005; Harris 2010). The ethical issues involved are beyond the scope of this paper with two unavoidable exceptions (but Sarkar [2021] provides a discussion and entry into the relevant literature). The first exception is that we will note, several times in passing, that many of the traits that are promoted have long and vicious histories of institutionalized racism associated with them—take, for instance, skin color and intelligence, both of which will be discussed in some detail below. The second exception is that we will explicitly note that social values strongly influence the selection of traits that would be slated for enhancement even though most eugenicists today emphasize that these decisions must be left entirely to individual parents.2

There is reason to believe that prospects for genetic enhancement are better for structural physical traits than they are for behavioral traits. At least a subset of physical traits has simpler causal pathways from the molecular to the organismic level than behavioral traits. Some of them are also clearly inherited even if patterns of inheritance are far more complex than those predicted from Mendel’s rules (indicating that a large number of loci may be implicated). We will use a suite of such traits that are often viewed as expressing societal values: pigmentation of eyes, hair, and skin. All three of these traits are known to be inherited to a considerable degree. All three traits depend on the presence of melanin, which comes in two types: a red-yellow form that is known as pheomelanin and a black-brown form known as eumelanin. Our question: How hard will genetic enhancement be for melanin-dependent traits?

Eye color, more specifically the color of the iris, is genetically the simplest of the three cases. Shortly after the emergence of genetics as a discipline at the turn of the twentieth century, Davenport and Davenport (1907) as well as Hurst (1908) claimed to have shown that eye color depends on one locus with two alleles.

Though this claim finds its way into textbooks even today, we now know that it is an over-simplification. (Exceptions began to be reported as early as 1909 [Sturm and Larsson 2009].) While Hurst recognized only blue and brown eyes, the phenotype is much more complex. As White and Rabago-Smith (2011, 5) have pointed out, “Eye color ranges include varying shades of brown, hazel, green, blue, gray, and in rare cases, violet and red.” The eugenic project is to reliably produce one of these colors in a lineage using CRISPR-based germline editing.

The physical basis for eye color is the distribution and content of melanocyte cells producing melanin in the front layer of the iris. An abundance of melanin would absorb a lot of light, resulting in a brown appearance. If there is less melanin, the color of the iris can be blue or gray or green in most cases depending on other molecular structures presents. (The other eye colors are more complicated.) The genetics of eye color is simple in the sense that, though sixteen loci have influence in regulating eye color, only two of them, HERC2 and OCA2, both located on chromosome 15, play major roles (Jablonski 2018, 1–2). These loci interact with each other in such a way that, contrary to traditional accounts, two blue-eyed parents can still give birth to a brown-eyed child (White and Rabago-Smith 2011, 6). This means, at the very least, that alleles at both loci will have to be edited simultaneously if we want to select a particular eye color. Another locus, MC1R, has been associated with green eyes and red hair though only in some populations—there will be more on this case below.

The trouble is that these two loci are not the only ones involved in eye color. In fact, genes for any of the several proteins that play a role in melanin formation and maturation also have a role in determining eye color as well as pigmentation of hair and skin. Genome-wide Association Studies (GWAS), designed to locate all loci statistically correlated with a trait and, thus, potentially playing a causal role in its etiology, implicate scores of yet other loci and different ones for different regions of the world (Maroñas et al. 2015). While what that means is far from clear, that totality of the evidence indicates that no small tractable set of genes have sufficient specificity for eye color to suggest successful genetic enhancement at the present state of our knowledge; however, enhancement may be possible for individuals with very well understood genomic backgrounds that have been carefully analyzed for the role played by HERC2 and OCA2. We do not have this kind of understanding for any population, and the uncertainties remain formidable.

Human hair color is a quantitative (i.e., continuously varying) trait that depends on the quantity, distribution, size, shape, and melanin content of the melanosomes, the organelles in cells responsible for the synthesis, storage, and transport of melanin. The most important factor is the ratio of eumelanin to pheomelanin. Maroñas et al. (2015, 16) note the following rules: Red and blonde hair generally contain less melanin than brown and black hair. However, blonde hair contains the same number of melanosomes as brown or black hair but have melanosomes that are smaller and rounder. Brown hair has large ellipsoid melanosomes containing mainly eumelanin. Black hair has the biggest melanosomes and most densely packed eumelanin molecules. Visual appearance is not a very good guide to the pigment composition of hair, and hair color changes in life not only in old age but also when a child grows into an adult.

Given these complexities, it should come as no surprise that a large number of loci are involved. Of the loci we have already encountered, MC1R is obviously relevant, but its variation seems to be important only in populations with a high frequency of red hair and fair skin. Some variants are associated with freckling. These traits seem to be associated with a high pheomelanin to eumelanin ratio (Barsh 2003). As expected, GWAS data have only complicated the issue. One study of a large sample of individuals of (self-identified) European origin identified 120 loci associated with skin color; another found more than two hundred (Pavan and Sturm 2019). The first study also found a sex bias in hair color with women having lighter hair than men. Given these genetic data, genetic intervention to change hair color also remains implausible at the present state of knowledge.

We finally turn to human skin color, which can vary from the darkest brown or black to the lightest white tones. Two types of skin color are usefully distinguished: constitutive and facultative (Maroñas et al. 2015). Constitutive skin color is the natural (“genetic”) color of the epidermis; facultative skin color is what it becomes under exposure to ultraviolet light (for instance, due to tanning) or to certain hormones (for instance, those present in skin-lightening creams). We will be concerned only with constitutive skin color. Light skin has a high proportion of light brown eumelanin and yellow/red pheomelanins in smaller less pigmented melanosomes that occur in bunches; dark skin has more dark brown eumelanin and larger densely pigmented melanosomes distributed singly.

Because of the association between skin pigmentation and racism, and the social privilege enjoyed by individuals in diverse cultures throughout the world, for instance, China, India, and Nigeria, besides the countries of the global north, efforts to lighten skin color are legion and include the routine use of skin-lightening creams (Khan 2018). Is genome editing a viable, more permanent, and even perhaps safer, alternative to skin-lightening creams? Folk biology would suggest so. Not only is skin color a quantitative trait, children typically have skin color intermediate to those of their parents; it is an almost perfect exemplar of “blending” inheritance. But folk biology is in for a surprise. The genetics of human skin color is so complicated that, at the beginning of the twenty-first century, Barsh (2003, 19) observed:

One of the most obvious phenotypes that distinguish members of our species, differences in skin pigmentation, is also one of the most enigmatic. There is a tremendous range of human skin color in which variation can be correlated with climates, continents, and/or cultures, yet we know very little about the underlying genetic architecture. Is the number of common skin color genes closer to five, 50, or 500? Do gain- and loss-of-function alleles for a small set of genes give rise to phenotypes at opposite ends of the pigmentary spectrum? Has the effect of natural selection on similar pigmentation phenotypes proceeded independently via similar pathways? And, finally, should we care about the genetics of human pigmentation if it is only skin-deep?

Fifteen years later, we are not doing much better. Leaving aside social issues, from a more purely medical perspective, we should care about human skin color insofar as different amounts on melanin can, on the one hand, influence susceptibility to ultraviolet radiation, which can lead to cancer and, on the other, the ability to synthesize vitamin D and prevent rickets. It turns out that the genetics of coloration is inordinately complicated. Though smaller numbers of genes have been implicated than, say, for eye color, different genes have been implicated for different populations. A 2007 study of South Asians found three genes—SLC24A5, SLC45A2, and TYR—to be most important in explaining variation in skin tone (Maroñas et al. 2015, 270). For a European population, the six genes implicated were HERC2, OCA2, IRF4, TYR, ASIP, and MC1R (Maroñas et al. 2015, 270); only TYR was shared with the South Asian population, but recall that HERC2 and OCA2 both play major roles in regulating eye color.

A study of African populations from Ethiopia, Tanzania, and Botswana implicated SLC24A5, HERC2, and OCA2, as well as three other genes (MFSD12, DDB1, and TMEM138) (Pavan and Sturm 2019, 58). A study of KhoeSan populations in Southern Africa found none of these genes important for skin color variation (Pavan and Sturm 2019, 59). A study of Latin populations implicated TYR, OCA2, HERC2, SLC24A5, SLC45A2, and IRF4, but also TYRP1 and a new gene not previously known to be associated with skin color (Pavan and Sturm 2019, 59). At the present time, genome editing would be no viable competition for skin-lightening creams.

3. When Claims Fall Apart: Complex Phenotypes and Behavioral Genetics

Let us turn to the genetic enhancement of complex traits potentially enabled by CRISPR technology and ask whether it is plausible or even possible. We will focus on a trait that has historically been very dear to eugenicists of every stripe: intelligence.

It will be useful to begin our discussions with the work of Plomin (e.g., Plomin and von Stumm 2018), who is one of the most vocal purveyors of the idea that genes are a major determinant of intelligence and, thereby, of the rest of intellectual and professional performance. Plomin was the primary source of a full-blown genetic determinism about intelligence that animated a leaked 2013 essay of over two hundred pages prepared by Dominic Cummings, a close advisor of Michael Gove, the British Secretary of State for Education from the Conservative Party. According to Cummings: “Work by one of the pioneers of behavioral genetics, Robert Plomin, has shown that most of the variation in performance of children in English schools is accounted for by within school factors (not between school factors), of which the largest factor is genes.”3 Those who endorsed Plomin’s views included Boris Johnson, then Mayor of London, who has since gone on to greater things.

One of Plomin’s modest proposals was the creation of a “genetically sensitive school” designed to match children with their presumed genetic capacities. Asbury and Plomin (2013, 366) presented this vision in a 2013 book, G Is for Genes: The Impact of Genetics on Education and Achievement:

We aim to treat all children with equal respect and provide them with equal opportunities, but we do not believe that all our pupils are the same. Children come in all shapes and sizes, with all sorts of talents and personalities. It’s time to use the lessons of behavioral genetics to create a school system that celebrates and encourages this wonderful diversity (187). . . .

One way of helping each and every child to fulfill their academic potential is to harness the lessons of genetic research . . . It’s time for educationalists and policy makers to sit down with geneticists to apply these findings to educational practice. It will make for better schools, thriving children, and, in the long run, a more fulfilled and effective population. That’s what we want schools and education to achieve, isn’t it? (3–4).

Asbury and Plomin’s vision for the school was truly breathtaking: “The site we choose for our genetically sensitive school will be enormous, more like a small university campus than a traditional school. It will have to be this size to hold all of the facilities it needs to accommodate and all of the options it needs to provide. It will serve the community around it, and we will make it so appealing and so successful, and we will foster such a pleasant environment and such a wonderful reputation, that every child of every faith, every race, and every social background will want to be educated there” (Asbury and Plomin 2013, 178–79).

Asbury and Plomin did not make an explicit case for genetic enhancement. They claimed the high ground by supposedly caring for and catering to all children, whatever their genetic endowments. But their discussion also makes it clear that genetic differences cause differences in achievement for a wide variety of skills, especially those like intelligence that carry social acclaim with them. Intentionally or not, they whet the appetite for liberal eugenics, that is, parental manipulation of genes to generate brighter children.

Plomin based his claims on behavioral genetics. So, we turn to that discipline to assess the plausibility of his claims. But, to start with, how does behavioral genetics choose what is a trait? For instance, is the number of teardrops shed in a lifetime a trait? If not, why not? Unfortunately, these questions do not have a technical answer within genetics; that is, there is no theory that determines what a trait is independent of what biologists choose to study for any reason (Sarkar 1998). For instance, biologists study features of organisms that appear to be functionally important (such as body weight, height, and a large variety of other morphological as well as physiological features). They also often study those features that appear striking, especially if they are also passed on from generation to generation (for instance, the number of fingers on a human hand or tumbling behavior in pigeons). Typically, they choose features with sufficient stability and regularity that they can carry out experiments with them, which is probably why the number of tears shed in a lifetime is not of much interest.

Plomin and others of his ilk simply assume that intelligence is a trait and that it is appropriately measured using IQ tests.4 Yet, psychologists such as Richardson (2000) have pointed to the remarkable malleability of what intelligence is supposed to mean. Mugny and Carugati (1989) carried out an experiment in which eight claims about intelligence were given to parents to grade as being correct with a score between one and seven. Many of these claims contradicted each other. Yet, many parents were content for all of them to be correct, thus endorsing contradictory claims about intelligence. For instance, there was equal agreement on the following two claims: that the “development of intelligence is the gradual learning of the rules of social life”; and that the “development of intelligence proceeds according to a biological program fixed at birth” (Richardson 2000). Another research group, Sternberg along with multiple collaborators, questioned a large number of psychologists about what they took intelligence to be. Of twenty-five possible attributes of intelligence, only three were mentioned by more than a fourth of these psychologists (Richardson 2000, 4). Richardson and many other psychologists have also pointed out the ideological role that intelligence plays in contemporary northern societies; it is used to judge people, especially children, and establish a supposed hierarchy of races. Mugny and Carugati (1989) note its cultural provenance: “Intelligence, if such a thing exists, is the historical creation of a particular culture, analogous to the notion of childhood” (quoted in Richardson 2000, 19). Sternberg is even more explicit: “Intelligence is invented . . . it is not any one thing. . . . Rather it is a complex mixture of ingredients. . . . The invention is a societal one” (quoted in Richardson 2000, 19).

Plomin and like-minded researchers claim to measure intelligence through IQ tests. Now, it is logically possible that, even if intelligence as a whole should be viewed as a collection of many different capacities, IQ scores capture one of these accurately (and IQ thus becomes a concept of some scientific value). So, we must turn to the question of what IQ is. Both critics and proponents of IQ are legion and some critical responses such as Gould’s (1981) Mismeasure of Man have achieved iconic status in the philosophy of biology. Critical race theorists such as Gillborn (2016) have persuasively argued that IQ has always been a tool of racial discrimination and subordination, no matter whether the targeted races were southern Europeans, Jews, Africans, or all people of color—in fact, the target has predictably varied with the dominant northern politics of the era. For instance, in the early 1920s, IQ testing of would-be immigrants at Ellis Island in New York City determined that 83 percent of Hungarians, 79 percent of Italians, and 87 percent of Russians were feebleminded (Richardson 2000, 30).

We must focus on how IQ tests are designed. The idea that there is a single general internal power or capacity for mental ability goes back to the nineteenth century, to Francis Galton, the founder of eugenics. However, his subsequent influence in the IQ story is negligible. That story begins around the turn of the twentieth century with Binet, who designed what has come to be regarded as the first IQ test to measure intellectual performance of children in Paris. Unlike those who followed his footsteps, Binet was not concerned with theorizing about the capacities he measured. Rather, his goal was to identify children who required remedial help in school irrespective of the causes. All that changed when IQ testing was exported to the United States, particularly in the hands of Terman, who was out to identify “mental defectives” for removal from society.

What matters most in our context is how Binet designed his test. For Binet, intellectual performance involved at least general knowledge, memory, imagination, attention, comprehension of sentences and synonyms, aesthetic judgments, moral judgments, a list far more comprehensive than the ones used by those who have followed him since (Richardson 2000, 26). To test for these capacities, Binet devised a vast array of questions that he proceeded to administer to children. Next, and this is the critical move, he selected a subset of these questions to be part of his final test using two criteria. The first was whether the average performance in answering a question became better with age; if so, it was supposed to give some indication of a child’s intelligence. The second criterion is the one that has characterized IQ tests ever since: Binet asked whether children’s performance on a question matched their teachers’ judgment of their intelligence.

Psychologists take it for granted that teachers are very good at predicting students’ academic performance. Given how Binet selected his questions to match teachers’ assessments of students, it follows that IQ can also correctly predict academic performance. Newer IQ tests, especially in the United States, followed Binet’s methodology and are typically referred to as Stanford–Binet tests (Richardson 2000, 31–36). The methodology has been manipulated by test designers in a variety of interesting ways. For instance, in 1937, it was discovered that girls on the average scored a few points lower than boys in the Stanford–Binet tests that were in vogue at the time. Test designers debated whether to allow this difference to persist and ultimately decided against it. The questions generating this difference were duly removed and we had a gender-neutral test by social construction. Other manipulations are equally informative. Tests always include a large number of questions that most people get right and very few that most get wrong or very few get right. The result: a bell curve for IQ—there is no more to that celebrated shape than an artifact of test construction.

So, where does this leave us? We are supposed to be measuring an invisible power, general intelligence, that some researchers have called g. What we get are visible data from IQ tests that consist of a set of scores. These visible scores are supposed to measure the invisible power, g. This situation is routine in science. Think of an invisible power called gravitation. The visible data are the positions of the sun and planets at different times. Using these data we can make many inferences about the power of gravitation, how strong it is, how it changes over large distances, and so on. But we can only make these inferences because we have a superb theory that connects the invisible to the visible. This is Newton’s theory of gravitation, one of the most successful scientific theories ever formulated (though it has had to be somewhat corrected by the general theory of relativity). But what about g and IQ scores? We have no theory at all.

It should come as no surprise to us that, beyond academic performance, most analyses show that IQ scores are poor predictors of job performance (Richardson 2000, 42–45). They are not even good predictors of cognitive ability, for instance, the mental capacities used by regular bettors at a racetrack (Ceci and Liker 1986). Then there is the “Flynn effect”: Mean IQ scores have risen around the world, year by year, decade by decade (Flynn 1987). In some countries, it has risen by fifteen points over three decades. All that is likely to be going on is increasing similarity between what test questions ask and what people are regularly doing more and more throughout the world because of globalization. People are probably not simply getting smarter, at least at that rate.

Does this mean that liberal eugenics is delusional if it expects to enhance intelligence through genetic manipulation? Well, Plomin and those of his ilk still have one card left to play. Even though many aspects of IQ may be artifactual for the reasons we have just seen, there may still remain an underlying core, g, that is stable and respectable because it has a biological, indeed, a genetic basis. In fact, the claim that IQ is genetic has been around for more than a century, though its popularity has waxed and waned. The claim has always been controversial. Historically, throughout the twentieth century, attributing IQ to genes became less popular when attempts to find the implicated “candidate” genes failed in spite of dedicated efforts; the claims became more popular when new technologies were invented to detect genes with supposedly more subtle effects on traits. Following this pattern, the launching of the HGP resulted in renewed claims that IQ is genetic (Sarkar 1998).

Plomin places his own work in that context as he promotes what he calls the “new genetics of intelligence” (Plomin and von Stumm 2018). These analyses rely on the latest technology for supposedly finding genes related to IQ, GWAS. Though most early attempts to use these studies to tie IQ to genes were unmitigated failures, according to Plomin and von Stumm (2018), recent results are supposed to have turned the situation around. What is at stake in these disputes is the concept of heritability, more specifically, broad heritability, H, of a trait. IQ is supposed to be genetic because it has a high heritability (of around 50 percent). The trouble is that the concept of heritability, and what (if anything) it shows about genetics, has been contested by geneticists ever since it was introduced in the 1940s (Sarkar 1998, chap. 4). Critics, and there are legions of them within the genetics community, hold that future generations will have as much truck with heritability as ours does with phrenology.

What heritability means will be central to our discussion of genetic enhancement of intelligence. But first, a point of clarification. Though what follows will be quite critical of the concept of heritability, we are not suggesting that genes have nothing to do with cognitive capacities. Rather, these capacities, like all other aspects of our biology, depend critically on our genes just as they do on the environment of development, that is, the history of interactions between genomes, cells, and higher-level factors. Returning to heritability, let us start with some well-defined trait, unlike IQ, that varies continuously across a population. We will use height.

We can measure the height of each member of a population. From these numbers we can calculate the mean and the variance. This variance is the phenotypic variance, $V_P$, of the trait (in our case, the height). To get to heritability, we must ask, what fraction of the phenotypic variability is due to genotypic variability (i.e., variation between individual genomes)? Since this variability can also be captured by a variance, and symbolized $V_G$, (broad) heritability can be defined as the ratio $H=V_G/V_P$, a number that can vary between 0 and 1. But how is $V_G$ to be measured? If we can manipulate experimental populations, we can at least approximately estimate $V_G$ For instance, suppose a population of plants is grown in such a way that all individuals experience the same environment. There are important subtleties here: We cannot grow all the plants in the same environment because they cannot all be grown on the same spot. Rather, we use everything we know about plant growth (ambient light and temperature, soil composition, pH, humidity, etc.) to make sure that each plant experiences the same value for all relevant environmental variables. (Already, there is an important disanalogy with IQ; even after all these decades of research, the exact environmental factors influencing IQ remain unknown.)

After controlling for environmental factors in this way, if we measure the variance in plant height, we have an estimate of $V_G$ for our experimental population and can calculate H. The trouble is that H, so estimated, depends critically on the genotypes in the population and the environmental values that the plants were grown in. If the genotypic composition of a population changes, as it would in the next generation unless every member is cloned, H would change; it would also change if the environment varies beyond what it was in the last generation (in our example, if it changes in any way).

If these aspects of H already suggest that its utility is very limited, there is more to come. Consider the human population in a landmine-infested region, for instance, parts of Cambodia in the 1990s. Now, presumably, we would all agree that all human beings have two legs for genetic reasons. Given that mutations that change the number of legs are very rare, for our Cambodian population, we would reasonably assume that $V_G$ ~approximately 0. Yet, there would be many people with only one leg in this population, because they would have had the misfortune of stepping on undetonated landmines. This means that $V_P>0$. We thus get H ~approximately 0 for a trait (having two legs) that should clearly be regarded as genetic.

Now, consider an example that goes the opposite way. Consider a population, and these are not hard to find in the United States, in which the only language spoken is English. Let us assume that there is small fraction of this population that does not speak English because of genetic cognitive impairment. This means that the only variability we have in the population is genetic. So we would have $V_P=V_G$ and $H=1$, even though the trait “speaks English” is not encoded in genes. Heritability is a strange measure indeed.

What matters most in our context is that what heritability measures is the fraction of the variability in a population. It says nothing about the mean value of the trait in the population, let alone the value it has in an individual. Thus, high heritability by itself says nothing about the genetics of a trait (as the earlier examples were intended to show). Consider height in humans. It is known to have a very high heritability, according to several analysis around 80 percent (McEvoy and Visscher 2009.) Going by how the heritability of IQ has been used by the proponents of a genetic basis for intelligence, the high heritability of height should be taken as evidence that the mean height of human populations cannot be changed except through genetic enhancement. Yet, the mean height of human populations throughout most areas of the world has been increasing each generation. Diets have had a tremendous influence. Heritability is a statistical measure, but it is not one that points toward a causal story.

Estimating heritability in human populations is non-trivial because ethics prevents us from setting up experimental populations. So, indirect methods are used, and these are fraught with problems that delight critics and are typically ignored by heritability enthusiasts (Sarkar 1998). One of the best-known of these methods is twin studies. A popular strategy is to study identical twins reared apart. For any trait when a pair of these twins shows differences, those differences supposedly can be attributed to the environment. Let us call the variance $V_E$. Now, if we assume that $V_G=V_P-V_E$, then we can estimate $H=V_G/V_P$ from studying these twins. Analyses of this sort have typically returned $H=0:5$ (or 50 percent). The trouble is that this calculation assumes that there is no interaction between the genes and the environment. But identical twins look alike, and, in most cases, that leads to other people treating them similarly. Thus, for cognitive traits, it is implausible that there ever is no interaction. There are other problems too: No analysis has ever convincingly shown that twins reared apart experience very different environments. Typically, adoption agencies try to match many aspects of the backgrounds. How important this is we don’t know; we have a very poor understanding of what environmental features are relevant for IQ.

This is where GWAS is supposed to have changed the game. It apparently provides an entirely new way of estimating heritability from the strength of the association of IQ with regions of the genome. Plomin and von Stumm (2018) review the evidence and claim that each new study since 2015 has given a higher value of $V_G$. But the best value they could report in 2018 is a heritability of 10 percent, which is well below the 50 percent estimate from the earlier problematic studies. For this 10 percent result, they report IQ to be associated with about a thousand different loci or regions in the genome. Even if we accept these results—and they remain controversial—this does not give much teeth to the claim that IQ is genetic. If thousands of loci can be involved, and each has a very tiny effect, it is hard to imagine a trait that would not be genetic in this rather toothless sense. Moreover, the problems with interpreting heritability remain, most importantly, how it provides no guide to the value of a trait has for an individual (or even for the mean value in a population). If only 10 percent of differences are of genetic origin, well, 90 percent must be due to environmental differences. The natural implication should be to focus on the environment—nutrition, exposure to dangerous substances such as lead and other contaminants in water, educational resources, etc.—rather than genetic enhancement.

Plomin is, of course, aware of the subtleties with heritability and why it cannot say anything about an individual person’s capacities. His response is to appeal to another counting technique that has emerged in the wake of GWAS: genome-wide polygenic scores (GPSs). Unlike heritabilities, Plomin and von Stumm (2018, 150) claim, “GPSs predict intelligence for each individual.” How is this supposed to work? Suppose there are n loci that are correlated with IQ scores and we have the numbers for these correlations. At each locus, we have a number of alleles that increase IQ. The simplest way to calculation a GPS is to multiply this number of alleles with the correlation value for that locus and add it all up for the n loci. This is what Plomin and von Stumm do. The trouble is that to view this number as the causal genetic contribution to intelligence requires a very vivid imagination.

For one thing, we are adding up weighted correlations but have no basis for regarding the correlations as relative causal contributions. For another, we are adding up alleged contributions of alleles and loci as if they never interact with each other. Genetically, this is fantasy. If alleles never interact with each other, we have no dominance, and whenever we look at well-studied genes some amount of dominance is all over the place. Moreover, we are concerned with hundreds of loci that are identified by GWAS as being associated with IQ. Each of these loci is supposed to have a very small effect on (i.e., a low correlation value for) IQ and also affect scores of other traits. Yet, for all the complexity, it is supposed to be the case that the different loci associated with IQ have no influence on each other. Plomin and von Stumm (2018, 156 [Box 7]) seem to acknowledge some of these problems when they later urge caution in interpreting GPSs.

It would take a very gullible eugenicist to believe all these claims about the new genetics of intelligence. But even such a eugenicist would be forced to admit that the prospects of genetic enhancement of intelligence remain woefully poor at the present time even with the availability of CRISPR technology. We would have to edit scores of genes simultaneously, with all the uncertainties associated with each of them, to hope for an increase of IQ by a few points. Because each of these genes would at best have a tiny effect on the desired trait while affecting many others, the process would not come close to achieving the requirement of specificity (as developed in the last section in the context of editing disease-implicated genes). At the very least, designing more intelligent babies through genetic intervention is beyond our capacities right now.

4. Polygenic Scores: Astrology for the Age of Genomics

As Rosenberg and collaborators (2019) have pointed out, polygenic scores come in two stripes, polygenic risk scores and genomic polygenic scores, as exemplified by Plomin’s GPS. The computation begins with GWAS as explained above. More specifically, as they explain:

Over the past 15 years, genomic analyses have identified thousands of genetic variants that contribute statistically to variation in complex phenotypes, traits that have complex patterns of inheritance and that are affected by large numbers of genes in combination with environmental factors. . . . In a typical genomic study of a complex human phenotype—a genome-wide association study (GWAS)—genotypes at thousands or millions of sites across the human genome are each tested in a sample of people for statistical association with the phenotype. Each variant identified by such a study as statistically associated with the phenotype can be assigned an effect size, representing the estimated magnitude of the increase in the trait (for quantitative phenotypes) or risk or liability for the trait (for binary phenotypes) that is associated with possession of a copy of the variant. (Rosenberg et al. 2019, 27)

The effect size is measured by the correlation coefficient. We believe it is an effect only because we assume that the causal influence goes from the genotype to the phenotype.

Next, these coefficients are aggregated as we saw in the case of GPS calculation: “For many complex phenotypes, identification and analysis of contributing genomic variants—most having small phenotypic effects—has led to the formulation of polygenic scores, quantities that seek to predict a trait value associated with a specific genomewide set of genotypes . . . For a quantitative phenotype, a polygenic score for an individual genome represents an aggregation, usually in the form of a sum, of the estimated effect sizes of the genetic variants in the genome” (Rosenberg et al. 2019, 27). This means that polygenic scores inherit all the interpretive difficulties that GWAS-generated coefficients have.

These difficulties are identical to those we encountered earlier in estimating classical heritability scores. GWAS scores depend on the population from which the individuals sampled are drawn. If the environments to which this population has been exposed change, for any complex trait, the phenotypic trait distribution would change and, therefore, so would the statistical association between trait values and individual alleles at each locus (i.e., GWAS scores). If the genotypic composition of the population changes, we can expect the same type of change in GWAS scores. The emphasis here on genotypes, rather than allele, is intentional; for the development of a phenotype, the role of an individual allele in a genome depends on what other alleles are also part of that genome, that is, the genotype. (The other alleles are part of the genetic environment of any given single locus.) The parallelism between these interpretive subtleties and those encountered for classical heritability scores is exceptionless.

When we turn to the aggregation of GWAS-generated association scores to generate genomic scores, the additivity problem returns, as in the case of estimating variances for heritability scores, and even worse. Plomin and those of his ilk claim that genomic scores can be used to predict individual phenotypic outcomes. They blandly assert the association scores (correlation coefficients) represent the quantitative values of causal contributions of loci without giving any reason for that assumption. Then they weight these using the number of alleles. In each case, population-wide parameters are used to predict what is supposed to be occurring during the development of an individual. Then they aggregate by adding up. Other than the fact that addition is venerated during this process, there is no biological justification for this procedure and no evidence that it correctly predicts phenotypic values for individuals. What we have is of as much intellectual credibility as predictive astrology, though now cast in the language of genomics.

5. Rethinking the Foundations: Final Remarks

The last section has made clear how bleak the prospects are for eugenics through genetic enhancement of complex behavioral traits. The focus was entirely on intelligence and that leaves open the possibility that it is an exception, and we would have better luck with other traits such as temperament. However, these other traits are even less biologically understood than intelligence in the sense that there have been fewer claims about their genetic basis. Those claims are no more sound than those made about intelligence that we just analyzed.

Mostly, there are GWAS results for a suite of psychological studies from schizophrenia to attention deficit hyperactivity disorder (Sarkar 2021). These GWAS results provide reason for increasing skepticism about how important genes are. When hundreds of loci are reported to be associated with a trait (or, typically, with variation in a trait), but each has a tiny non-specific effect, knowing only these genes tells us very little about causal pathways, how an organism ends up with a trait during the course of its life. We must know how these genes are used, in what sequence, and what environmental factors must be using these genes.

In other words, we must understand the contextual developmental construction of an organism. Genes are no doubt important; each of us would not be who we are without the set of genes we have inherited. But we are as much the product of our development, initially within the womb, when many of our biological features were built out of environmental stimuli, and our continued post-natal development during which a multitude of cultural determinants—from nutrition and parental care to physical activity and intellectual stimulation—set the capacities with which we enter adulthood. It is pointless to ask whether nature or nurture (interpreted as genes or environment) is more important, and worse than pointless to say that the relative contributions can be quantified, especially by mere statistical associations. That is the main takeaway from the critical analysis of heritability and GWAS (Sarkar 1998, 2021). Instead we should recognize, along with Hogben (1933), the “interpenetration” of nature and nurture.

Biology is a part of what we are, but only a part, and that to a different extent for each aspect of who we are. Moreover, our genes are only a part of our biology. The genes that may be relevant to us many not even be our human genes; rather, the genes of the microbiota within and on us may be critical to our development. Some of our psychological features depend on our microbiome (Sariola and Gilbert 2020). Even more importantly, our senses of personal identity, who we take ourselves to be, is culturally mediated to such an extent that we cannot even cogently maintain a distinction between a biological base and cultural superstructure.5 The consequences of changing genes for psychological (and other complex) phenotypes cannot be specified by biological reasoning alone.

Returning to genetics, what we learned during the twentieth century was that single loci control a tiny minority of traits in most complex organisms (i.e., those that are relatively large and have some level of cellular differentiation), and the prevalence of these traits decreases with organismic complexities. The simplest traits are the one on which twentieth-century geneticists focused because they were tractable (Sarkar 1998). Because of this focus, these are the traits that have received most attention, whether it be in pedagogical contexts or by the general public. This attention has resulted in a misleading narrative of genetic dominance over the rest of biology. That is what has spurred eugenic programs such as genetic enhancement.

The preponderance of traits does not have simple Mendelian patterns of inheritance because these traits are influenced by a multiplicity of loci interacting with a wide array of environmental factors, as we saw for both physical (structural) and behavioral traits. So, it should come as no surprise that the whole genome DNA sequences produced by the HGP and the techniques it spawned had very little predictive power compared to what had been promised by proponents of the HGP (Sarkar 2002). GWAS results underscore this point. So does the rise of epigenetics, which was, in a sense, made necessary by attempts to formulate a predictive biology at the molecular level given how little DNA sequences alone contributed.

In this context, positive eugenics through genetic enhancement seems little more than incompetent science fiction irrespective of the ethics of enhancement. Changing our offspring’s genes has little chance of changing their personalities and capabilities or, very likely, even their physical features in the ways we may fancy, and only in those ways. What the last sections of this paper have argued is that this is not a question of technological proficiency; it is a question of fundamental biology. If we want offspring of a particular sort, our best bet is to raise them to have the capacities to be of that sort, as much as possible under the constraints of their biological constitutions. The rest is for them to decide and assert with the freedoms due to every human being.

Notes

Parts of this paper are excerpted from Sarkar (2021). Thanks are due to Christopher Donohue and Alan Love for comments on an earlier draft.

1. 1. This concept of specificity, which can be traced back to Timoféeff-Ressovsky and Timoféeff-Ressovsky (1926), is developed in detail in Sarkar (2021).

2. 2. This is the program of liberal eugenics. Sarkar (2021) provides a discussion that also contrasts it with moderate eugenics, which tolerates some degree of societal intervention in reproductive choices.

3. 3. Gillborn (2016, 368). Much of the details of these episodes are from this source.

4. 4. Liberal eugenicists do not question this assumption.

5. 5. Note the reinterpretation (and rejection) of an old Marxist distinction.

References